Everpure (PSTG) Earnings Call Transcript & Summary

Okay. So hopefully, I counted 321 correctly. So warm welcome, everyone, for our fireside chat with the team from Pure Storage. Before we get started, I would like to read the disclosure. Statements made in these discussions, which are not a statement of historical fact are forward-looking statements based upon current expectations. Actual results could differ materially from those projected due to a number of factors, including those referenced in Pure's most recent SEC filings on Forms 10-Q, 10-K and 8-K. So to that end, I want to thank everyone for joining us, a perfect timing given for Pure's recent earnings call.

I do have some questions, kind of follow-up, and I'm just going to start. Rob, you guys talked about hyperscaler revenue that is embedded in your fiscal year guide, especially for the second half. What I want to better understand is the diversity of customer. Is that just a one hyperscaler customer, too? And then the mix, does it include both hardware and software?

Yes. Thanks for the question, Mehdi, and thanks for having us, as you mentioned, coming off of a great 4Q print and had a lot of discussions with the financial community over the last couple of days. So let me hit your first question, which is, I think, just seeking to better understand the hyperscaler revenue expectations, where is that coming from? How does that look? And then a couple of elements in what we introduced as the new more standardized hyperscaler business model. As you mentioned, we discussed on the call on Wednesday, our increased expectations for the year relative to our prior commentary in terms of hyperscaler shipments, that's -- as we have tried to share in qualitative color, that relationship continues to go very well and is ramping quickly and has surpassed our prior expectations. We do expect most of those shipments in FY '27 to occur in the back half, so in the fiscal 3Q and 4Q just due to the schedule of the hyperscaler build-outs. To your question about customer mix and concentration, yes, at this point, we are still shipping to our lead customer in the space. We are engaged with -- continue to be engaged with in advancing engineering discussions and I would say, exploration and testing with additional future prospects, but really just one primary customer in the scaled part of the hyperscaler business at this point. And then I think the other part of your question, and I know I'm going along here, but you had 3 questions baked in. I think the other part of your question is seeking to better understand the mix of that revenue in terms of, hey, how much is software, how much is hardware? What does that exactly look like? So I want to take you back to and take your listeners back to our original business model, which we articulated in fiscal '26. And in our 3Q call, we had kind of foreshadowed and indicated, hey, we're going to be evolving this business model and then take you to what that looks like and our expectation moving forward and what would cover most of the FY '27 shipments. So the original business model, if you recall, was really centered around -- think of it as a software royalty or license fee for the direct flash software. That is really the heart of the IP. That's the heart of what the hyperscalers are interested in. And in FY '26, that was -- that made up the bulk of the revenue. A smaller piece in terms of maintenance and support, but we can set that aside. Really, the bulk of that was in software royalties. The hyperscaler supply chains would then in order to take advantage of that software, hyperscaler supply chain would be procuring the servers, the DRAM, the NAND itself, all the components to make the NAND into the direct flash modules, et cetera. What we learned as we've grown the relationship and have talked to additional hyperscalers is hyperscaler supply chains are quite adept at procuring servers in DRAM, quite adept at procuring NAND and NAND flash. For all of the other components to finish out the direct flash module, our PCBs, our designs on flash controllers, et cetera, it just made sense for all involved. It was a lot operationally simpler for our supply chain to go help procure those components. To be very clear, and we discussed this in Tarek's prepared remarks, we would not be procuring the NAND for the hyperscalers or the servers, really the rest of the components to make that solution. And so when you consider that in the overall shipment, that is what -- that change or that shift to that standardized business model is what is now bringing the gross margin expectations associated with the hyperscaler business into that range of 75 to 85 points. And depending on the configuration or the performance tier or the drive sizes, you may have some variation within that range, but that would be our expectation for that hyperscaler business.

That's pretty, I would say, clever, let your customers and their scale be involved with NAND procurement, which hasn't really happened in the past. I guess, perhaps in the past, most of the demand was driven by enterprise. This is -- the AI has enabled you to penetrate hyperscalers, and now they are helping you with procuring the required NAND. Would you agree with that?

Well, so yes, let me maybe unpack that. So obviously, there's a lot of complexity that goes into sharing and coordinating supply chains. And as you mentioned, we -- for our enterprise customers for our core business -- our customers are just looking for the best solution, finished product, et cetera. So it really doesn't make sense and they don't have the infrastructure in place to go run these types of quite sophisticated supply chains. Hyperscalers, on the other hand, it makes sense to -- for us to work together with them to apply the best capabilities they have in their supply chain, the best capabilities we have. And that's really what led to this evolution and standardization of the business model.

Yes. Now I want to -- just about of the queue I have with my questions and topic. And actually, go to the demand side, and there are evolving different drivers. How do you see KV caching impacting the demand, especially as we migrate to inferencing? And what are the key products in Pure's portfolio that will be applicable to that, which I would phrase as a new kind of a demand driver, the KV caching?

Yes. So if I step back and look at the -- now shifting gears to the core business, if I look at the demand drivers on the core business, I think we see a number of demand drivers we talked about on the call and reflected in our guidance and outlook for the year. We see robust demand across the board and exiting fiscal '26 with a ton of momentum. And I think there's a number of demand drivers across the industry. And then I'll answer your question specific to AI, which, of course, is still a smaller part of our overall book of business. I think that certainly if I look across the industry, what we saw in -- throughout the back half of fiscal '26 you saw us raise and subsequently beat the guidance several times through our fiscal '26. What we saw was a couple of things. One was a pretty robust demand across the board across segments, whether that was enterprise, commercial, U.S. international, it was pretty broad-based. And I can't identify specific application driver or specific vertical or area. We saw pretty broad-based demand, and you saw that come through in terms of our sales and our revenues as well. I think I suspect part of this may also be last year, right, we had a little bit of, call it, a headwinds to IT budgets across the board, perhaps exacerbated by the Broadcom VMware action. A lot of folks were pretty stretched, and you see some of that either pressure relieving, if you will, or maybe folks that had sweat assets a little bit longer. You see some of that starting to kind of relieve. I think the other thing is, as folks now turning more in the direction of AI, I think as folks are progressing in their thinking and plans about, hey, how do I go get more value out of my data modernization -- infrastructure modernization projects overall. We're seeing those move forward. And we see that both in the core array business, array and blade business as well as what we're seeing in Portworx in terms of modernizing not just container applications but virtualization as well. If I look specifically at AI coming to your question about KV cache, we're engaged in conversations across the board, everywhere from folks that are just looking to connect their data to chatGPT all the way to folks that are doing some small-scale maybe fine-tuning, maybe folks that are doing -- really focused on inference and RAG and to large-scale GPU clouds or foundation AI natives that are really down the training path. I think we have a robust portfolio of solutions for that. On the inference side, we do see more of a focus on KV cache. I think it's very early days. We have a number of customers on our FlashBlade solution that are making use of our object capabilities and our KV cache accelerator there to really make that whole process a lot faster. I do think as inference grows to scale, that will become more and more important. But as a portion of the market and just overall segment of kind of a share of our book of business, it's still relatively small compared to everything else that we do.

Yes. Yes, especially your NAND suppliers are having a hard time sizing the market -- the size of the market for KV cache and so everybody is trying to figure this out. But it's very fascinating because you have the enterprise finally beginning to utilize AI and then you have the native data, the hyperscalers wanting to own everything. They have already invested billions -- hundreds of billions in training and now they want to extend their services to inferencing. So in that context, how do you see neocloud service providers that are telling enterprises, hey, I could be your edge compute evolving. When you say enterprise demand is strengthening, is that traditional enterprise? Or does that also include neocloud service providers?

Yes. So a couple of pieces to the question there. When we talk about demand strengthening and just being robust across the board, really talking about traditional enterprises, traditional commercial, public sector, really our core business. And again, just highlight that, that we do see that pretty robust across all workloads, of which AI is just one of many. In terms of how our -- where do neoclouds fit into the equation? It's interesting, they -- I think we work with many neoclouds. We work with them on behalf of both, call it, large-scale tech titan types who are utilizing their capabilities as well as smaller enterprises and commercial customers that are utilizing them to provide the GPU infrastructure. I think the neoclouds are really focused on -- they're focused on providing infrastructure that can meet the wide range of demands. Think about it this way. If you're a GPU as a service provider, a neocloud, you might have one customer who's doing super large-scale foundation model training. You might have a couple of other customers that are just doing inferencing. You might have a couple of other customers that are actually working with images and video instead of text. So the range and variety of different needs that you have to go satisfy is quite large. And I think that presents an opportunity for us because unlike most of the storage competition in the AI space, we have not taken -- we've purposely not taken a laser-focused view, right? We've leveraged, we've really leaned into the power of our core technology, which is to be very flexible, very adaptable, very deployable for multiple different needs. And I think we're seeing that value proposition resonate as we engage with more neoclouds.

Yes. I agree. And I'm going to ask you one more open-ended question. I think there are still some details to be worked out, data sovereignty, data security, and I think your 1touch acquisition announced the other night fits into that narrative. But ultimately, I'm looking for an inflection point. We love inflection point where one earning report, you come out with a with a big revenue growth exceeding 20%. Are we -- how should I think about that context? Are we setting up to the -- over the next couple of quarters where you're going to see a scaling of your products? We're finally transitioning into deploying AI among enterprises. Or is this going to be just a slow train moving forward, but it's just not -- is this slow train is not going to turn into a bullet train? Hopefully, that makes sense to your question.

Well, I mean, I would just point out that the revenue guidance and outlook we provided for FY '27 almost doubled our initial outlook coming into growth coming into FY '26. So I would argue that we are driving significant acceleration across both the core business as well as our newer markets, whether that's the hyperscalers, whether it's AI. Look, we're focused on continuing to drive growth, continuing to accelerate. We have multiple levers that we're pulling on. And just to reiterate for your audience, certainly, the core business focused on completing the portfolio -- well, we've completed the portfolio focused on layering on top of that, the intelligent control plane and data management capabilities, things that will allow us to reach into -- and hire into larger accounts, compete for larger wallet share. And you see the results of that already starting to pay bear fruit with -- I think, Tarek shared a number. The number of deals that we track and have booked over $5 million, growing 80% year-over-year that, again, gives you a data point in terms of the leverage we're driving in the core enterprise. Hyperscalers, we've talked a lot about that in terms of the tremendous opportunity there. And notwithstanding, we're in a bit of a tough position to give too much disclosure because of the customer concentration. But needless to say that business is growing ahead of our expectations, and we've tried to share some color on that. And then certainly, you and I have had a lot of discussion around AI and the number of different angles that Pure -- I guess, Everpure, I should say, is playing in AI, whether that's in the enterprise space with KV cache and inference, whether that's really reaching deeper into the neoclouds with FlashBlade exit. So we've got a number of levers. We're focused on sustainable growth and accelerating that over time more so than individual inflection points.

Yes. And I think your execution over the past several years is a proven point with how well you're situated with enterprises. Perhaps it's the hyperscalers that if I were to use a scenario analysis could provide a growth acceleration. Would you agree?

Well, I think that any of those levers could provide growth acceleration. And we're looking to drive acceleration out of out of all of those. But yes, I mean, just by kind of the size of the market and so on, I mean the hyperscalers, it has dominated our conversations, is a huge focus area for us, and it's one of the big drivers that we're really leaning into.

Do you feel like you're in a situation where you want to drive operating growth profit? Do you want to drive operating margin expansion, but you also have all of these opportunities? So how do you manage the OpEx, right? Because you can allocate more OpEx to drive growth acceleration. But then you do give up some margin. Am I thinking about this right? You have to deal with a delicate balance, right?

Yes. So [ Leah ], let me share kind of our views on where we're focused. I outlined and we discussed all of the exciting kind of areas that we see as capable of driving growth, the levers that we're pulling on to get there. So how do we then think about resource allocation and how do we think about driving growth versus profitability? And the short answer is our thoughts and philosophy on that really haven't shifted. Charlie, Tarek, myself, the entire management team, we are -- make no mistake, we are focused on driving growth. We're focused on leaning into those levers, but at the same time, continuing to drive moderate leverage over time. Where we've tried to quantify that, just kind of roughly for the Street is we want to drive top line growth while every year accreting roughly speaking, a point to the OM. That said, and I think what Charlie -- what Tarek has articulated as he's joined is the better metric and where we're really more focused on is driving OP or operating profit dollar growth. Now obviously, you can do the math and back into what the implied OM is. But that's kind of how we're thinking about the balance of investments, growth opportunities, but then also continuing to drive incremental improvements to profitability.

Does the 1touch acquisition help you with hyperscaler opportunity, enterprise or both? And that was very interesting. My interpretation of 1touch acquisition is that, yes, it does help you with streamlining the data solvently, data security. So maybe you could just quickly explain what 1touch acquisition is all about? And how is it going to impact enterprise, hyperscalers or both?

Yes. Happy to. So let me hit the first thing -- the second thing first. It's primarily targeted. We see it primarily targeted towards the enterprise and the parts of our portfolio that exists today, but also, we will build in the future targeted at the enterprise customer base. What is 1touch? What is the strategic fit? I talked a little bit about this, or we did on the conference call. But if I step back for a minute, they really provide a couple of critical capabilities that we will both continue to sell in their existing offering, but also integrate into our core technology over time. These critical capabilities are the ones that come to the top of the list our data discovery, being able to automate data source discovery, data classification, right, being able to look at different pools of data and then generate from that and understanding of, okay, what is this? Is this PII? Is this customer data? Is this health records, et cetera, et cetera? And then the third is the ability to contextualize that, the ability to represent and capture semantic meaning of the data. If you net all of that out and you look at what we have been articulating as our Enterprise Data Cloud vision, our intent is to integrate those components into Purity, into Fusion, into the Enterprise Data Cloud over time so that we can go help customers better understand the meaning of their data across multiple different sources, multiple different silos and work with their data much more seamlessly. Charlie has talked about enterprise data is still trapped, is still subservient applications. We do a great job of providing the infrastructure for that today, where 1touch and the integration of those capabilities will come in and help is it will help customers actually be able to integrate data across multiple sources without having to do the old school thing of ETL and copying data all over the place. That said, we just announced the intent to acquire, we have yet to close the transaction. We expect that to happen at the beginning of our Q2. And then as I was telling somebody the other day, then the work began. So then we begin with the technology integration. But very excited about it. I think the strategic fit is quite exceptional.

And this goes back to what Charlie referenced earlier in the call, the unified control plane, and I'm not sure how many people picked that up, but when we think about unified control plane and given where the data lakes located in various places, the 1touch helps consolidate and actually make the unified control plane more effective, right?

Yes. I think it complements it very nicely. Think of it as the unified control plane helps customers consolidate infrastructure, where 1touch integration will help is they'll help customers consolidate the logical data and the use of data. So it's very complementary.

Yes. And what's also fascinating, I'm just making an observation here, enterprises wanting to do inferencing or deploy the infrastructure for inferencing on their own, given the data security and everything that you already mentioned. So -- and I do get a sense that maybe Pure is also strategizing to capitalize on this. So how do you deal with the actual deployment of necessary server rack? You are still a point solution, but you offer a comprehensive storage portfolio. Is that your enterprise customer that has to think about deploying the compute and storage together? Or is there like a third party that is involved? Any -- am I thinking about this right? Feel free to correct me what I just laid out.

Yes. So I think the essence of your question is, hey, we're a great data storage infrastructure provider. I'm talking a lot about additional software. How does the customer go and deploy that? We don't ship servers. What does that look like? So look, we work very closely with, I would say, converged infrastructure partners, Cisco being probably the strongest one, where we do have converged -- everything from converged solutions and offerings to strong reference architectures with multiple server vendors. And of course, our strong partner in GSI and systems integrator community works to help with deployment for customers. What I'd say is that especially with modern applications, the shift to cloud native, really Kubernetes and container-based deployment of applications, in some ways, has made this part of the puzzle a lot easier to fill in. Frankly, the value and where customers are focused is how do I deploy containers and Kubernetes quickly. In most enterprise customers, they're getting to the point where if you can walk in and say, hey, I've got the software offering, it's container and Kubernetes deployable. They've got a Kubernetes team somewhere that knows how to go do that. In some sense, the specific servers and racking and stacking that really has become commodity. And so it's not an area that we're particularly focused on.

And do you have any efforts to work with different CPU architecture? There's also competition on who is actually -- which kind of a compute architecture would do the inferencing, the XPUs and CPU and so forth. It's a diverse set of technologies out there, right?

Yes, there's a diverse set of technologies. We work with -- and really our customers deploy solutions on just about all the major CPU vendors, major and less, I would say, smaller GPU providers as well and everything in between. Clearly, in the GPU space, the bulk majority of our customers, whether it's training or inference, are running on NVIDIA, whether it's NVIDIA GPUs in third-party servers or NVIDIA provided servers. NVIDIA continues to be an incredibly strong partner for us and a focus -- really a focal point of our AI solutions development, as you might imagine.

Sorry, there's a fire drill going on. Apologies for background noise.

A literal fire drill.

Yes. I want to pivot back to the supply chain. And focusing on the enterprise side because on the hyperscalers, they're doing -- they're helping you with procurement. But on the enterprise side, how much of a visibility do you have with the demand? Is there like a backlog? Because you have to turn around and think about procuring the NAND, especially the tightness that may not go away. So help us understand the contracts you have with the NAND suppliers for enterprise application? And how are -- is there any flexibility to adjust the shipment? And how does the pricing kick in here? How often does the pricing is negotiated?

Yes. I mean -- so what I'd say is that we're -- our supply chain team for years, has been top notch. You saw the incredible work that they did to position us well through -- to weather the post-COVID supply chain tightness in calendar '21. The nature -- I'm not going to get into specific contracts and kind of numbers, but the nature of our strategic agreements and long-term agreements with our critical suppliers hasn't changed. If anything, that's really strengthened over time. We, as you might imagine, have multifaceted sets of agreements that, at the end of the day, give us very good visibility into our enterprise supply. I will say, and Charlie touched on this on the call as well, there are many components that go into a finished solution. The -- I would say the tightness on the overall supply chain situation right now is not just constrained to NAND, as you would know, covering the industry, it stretches everything from NAND to hard disk drives to HBM, high-bandwidth memory to DRAM to CPUs to networking cards. So our teams are very hard at work and busy ensuring that we have the supply we need to meet the enterprise demand that, as I mentioned before, is quite robust. We do benefit from the same structural advantages that I think helped us through the post-COVID tightness. And for your audience, just as a reminder, because of our software efficiencies, we're able to use far less components to the order of 20% to 30% than a lot of our enterprise competition. The fact that we have one unified hardware technology means that our overall BOM or bill of materials is a lot lower. There's fewer different parts to go chase. I have a lot more flexibility in steering mix in terms of a lot of shared parts means that for whatever I source, I have more flexibility to build all the things in my portfolio. And so these are things that I think we're also leaning into.

Got it. Two quick follow-ups. Can you update us on the mix of the QLC NAND that you're using? And how does that mix changes into the second half of the year?

So yes, I mean, we've been shipping more QLC than TLC as a business in the enterprise, so setting aside the hyperscalers for quite some time. I think that mix continues to shift more and more towards QLC, but it's been over 50-50 for quite some time. I haven't looked specifically first half, second half, but the mid- to long-term trend is pretty clear. It's kind of the QLC portion is growing.

Yes. And to what extent are you able to pass on the incremental cost increase to your enterprise customers?

So I think -- and the best way to look at that is the commentary we gave on gross margin implications, which is that -- which is to say we do expect near-term pressure in Q1 gross margins on the product side, specifically because the input costs grew very, very rapidly in the back half of Q4 faster than the ability for the market pricing to adjust. We've since taken pricing action. We do expect the effects of that pricing action to really help us recover those gross margins starting in Q2 and then throughout the year. That -- all of that said, we don't -- to be very clear and to restate, we don't we don't think of our pricing philosophy and our pricing actions as being gross margin driven. We look to the market. We price in a competitive space. We command a premium for our solutions. That's reflected in our long-term gross margin benefits over the competition. What happened here was that, again, the rate of input cost increases was almost unprecedented faster than the market could adjust pricing. And so what we've done in those pricing actions is we've reevaluated, reassessed where market pricing has moved to accommodate the new higher level of input costs. We've adjusted our pricing, and that's really what you're going to see starting to flow through Q2, Q3, Q4 so that the gross margins recover back into a healthy part of that long-term range of where we want to be in the product side, 65 to 70 points.

Great. Okay. So we're a minute over the time. I want to thank you for giving us opportunity to catch up with you post earnings. I want to thank everyone on this call for participating. If there's any follow-up from investors, feel free to send me an e-mail, [email protected]. I wish everyone a great weekend. Thank you, Rob, and the rest of the IR team.

Thank you.

Thanks, Mehdi.

Bye.